Phillips-Ouliaris Cointegration Test

[irulishka]

Hello,
Could you explain why in Pillips-Ouliaris cointegration test with d.f. adjustment to estimate “Bias corrected Rho - 1 (Rho* - 1)” you use not d.f. adjusted Long-run residual autocovariance? Workfile and excel data attached, selses F21:H21

[EViews Glenn]

Not sure. We'll have someone take a look. We're a bit swamped right now so it may take a bit. It 's possible the option isn't getting passed down properly.

[EViews Glenn]

The PO test should not have an option for specifying a d.f. correction on the long-run variance (which is why it appeared it was being ignored). There are several conceptual issues about how to align various d.f. corrections in the PP test on residuals from a cointegrating equation (since there are two equations, the long-run regression and the unit root regression and rather than trying to mix and match various cases to ensure that things are done consistently, we decided to not perform d.f. corrections when constructing the test statistics (Z_alpha and Z_t) .

If you requested a d.f. correction, the reported intermediate test statistics did apply a d.f. correction but only for the unit root regression portion of the test, not for the first-stage regression. This correction was applied equally to the reported intermediate variance, long-run variance, and one-sided long-run variance. But the uncorrected values were used in constructing the statistic.

So all of the calculations are correct, but since I think this is potentially confusing, it is best if we disallow the d.f. correction entirely. I've gone into the code and removed the option. Note that no final statistic values will change, but it is possible that some of the intermediate results will change slightly.

I hope that this answers you question. Thanks for pointing out the discrepancy.